The aim of the project is to test the effectiveness of machine learning methods in predicting active sites on proteins from their primary structure.
Although methods utilizing 3-D models of molecules are more effective in predicting active sites on proteins, methods based on primary structure are still useful. The disadvantage of these methods is the absence of a 3-D model for most proteins. Therefore, it is beneficial to have a tool that can predict active sites solely from the sequence of amino acids composing the given molecule.
The program takes a sequence in FASTA format as input and classifies individual amino acids as one if they are expected to be part of an active site, or as zero otherwise.
The goal is to prepare data that will be used to train the program.
- Data loading: The data is obtained from https://github.com/rdk/p2rank-datasets. The P2RANK tool (https://github.com/rdk/p2rank) is used to extract primary structures of proteins and their known active sites from the data.
- Assignment of active sites to each sequence and splitting into individual chains.
- Data validation: Checking the processed data.
- Trying different machine learning methods for predicting active sites.
- Comparing the effectiveness of different methods.
- Comparing with P2RANK.
- In folder trainData there are data used for training machine learning algorithms and description of how to regenerete them.
- In folder pbsDestription there is code used for testing machine learning methods.